It seems we don't actually close() remote sockets, we let the destructor
do that.

Also, log it when it's changed.

This may be the root cause of the extended netsplits we've been seeing.

* Support Unreal-style AKILL messages by way of TKL.  Propagate to all
  other servers on the network.  Thanks to Netsurge for bringing this to
  my attention.
* Propagate AKILL messages before local kill testing is done.
* Some minor efficiency and cosmetic changes as per jslint.

  first user to create the channel.  This comes on a MODE message later.
  Thus, don't assume the first user in a newly created channel is +o unless
  a local user.  Bug found by wb3lli, thanks!

  that as a valid BBS login.
* Update copyright to 2008.

* Setting +l with an integer greater than 99999 would result in +l being set
  without any argument.
* Better handling of TSMODE from DreamForge servers.

  (randomly) swap the 'hops' and 'created' variables on the NICK line without
  warning.  The IRCd now tests for this.  This manifested itself with time_t
  integers appearing in the 'WHO' command output when a CR server was linked.
* WHOWAS output would always show the uplink server for a client who left the
  network from *behind* the uplink.  The correct server info is now shown on
  WHOWAS.

* Minor rework of how connected clients and servers pass commands.

  the built-in JS "Array" object.  This is because the Array object offers
  no facilities for associative arrays.
  Full details here:
  Thanks to Deuce for pointing this one out.  This fixes many minor quirks,
  such as "undefined" coming up in STATS M and soforth, since a general scan
  of the array object would also pick up on properties of the Array object.
* More general code cleanup, especially in ircd_server.js

* Scanning of QLines now uses wildmatch()
* Fixed a crash involving WHO +C
* Lots of general code cleanup
* Wrote in profiling support, finally!  STATS M will now give a list of
  executed commands, the total clock ticks per command, and number of
  executions per command.

* irclib IRC_string() function now behaves exactly like every other IRCd in
  the world, and requires the "arg" argument, which defines which word to
  begin scanning at in the string.
* irclib IRC_match() is now obsolete thanks to DigitalMan's wildmatch()
  function in Synchronet 3.14 -- all IRC_match() functions in the IRCd now
  switched to wildmatch()
* More general cleanups and optimizations.

  BBSnet to test DreamForge<->Bahamut translating.  Anyone can test this out
  by connecting to rrx.ca, port 7777.

  attached to the receiving server (as opposed to the server string itself)
  would crash the IRCd.
* Beginnings of Dreamforge support.

  Thanks to Tracker1 and RuneMaster for pointing this one out.

  as expected (leaving dangling +l modes without an integer)
* Updated copyright

  would cause a server with a wildcard C:Line to continue to attempt to
  connect to its uplink even after a connection was established.
* When connecting to another server, it was possible that the IRCd would
  send its PASS/SERVER command pair again in error.  This would generate
  warning messages on the receiving server side stating "Refusing to comply
  with supposedly bogus SERVER command."
* When a channel was both +k and +l, the IRCd would incorrectly interpret
  the incoming SJOIN message for that channel.  (i.e. +kl 1 1 would return
  +k 11)
* The IRCd now crunches modes on an incoming SJOIN message.

filtering by log level by sysop configuration.